Market volatility and performance of ESG portfolios

(1)

Market volatility and performance of ESG portfolios

Abstract

In this paper we examine the relation between market volatility and performance of ESG portfolios between 2002 and 2017. North American and European firms are placed in separate data sets covering, on average per month, at least 1000 and 750 firms respectively. These data sets are used to create market value weighted and equally weighted portfolios that include only the top 2.5%, 5% and 10% of ESG score firms, respectively so for the bottom ranked firms. We construct additional portfolios with long positions in portfolios with top ESG score firms and short positions in portfolios with bottom ESG score firms. Multiple regression analysis reveals that North American top ESG score portfolios have a significant positive relation to their volatility index and that European top ESG portfolios have a significant negative relation to their volatility index. However, we do not find a consistent significant relation between returns of bottom ESG score portfolios or returns of long/short ESG score portfolios and their respective volatility indices.

Student: S1893068

Name: Richard Nijensteen

Program: MSc Finance

Supervisor: Dr. L. Dam

(2)

1. Introduction

Environmental, Social and Governance (ESG) investing, often referred to as socially responsible investing (SRI), has been on the rise in financial markets for decades. According to the US SIF Foundation, since they started measuring socially responsible assets in the US at 639 billion dollars in 1995, the amount of SRI assets has grown to 8.7 trillion dollars in 2016 and 12 trillion dollars in 20181_{. In order to guide investors as they increase exposure to these} assets, we need to be able to explain the return on portfolios within the realm of ESG investing. An abundance of characteristics of ESG portfolio return has been explored in the field of finance already, but research in the interaction between volatility and ESG asset return remains underrepresented.

Currently, ESG assets are primarily held by one of two types of investors. The first are those who are committed to expressing their personal values by investing solely in responsible firms, where a firm acts as a channel for these investors their values to bloom, coined “Delegated philanthropy” by Bénabou & Tirole (2010). The others, a much larger group including institutional investors, do wish to support these firms, but only at a very limited cost (Desclée et al., 2016). Institutional investors especially will benefit from further research into the factors that constitute ESG portfolio returns by better means of estimating the implied risk of their investment funds.

Thus, we form this paper around the following research question:

Does the volatility of a market explain the return of portfolios restricted by aggregate Environmental, Social and Governance scores?

In order to create distinct portfolios that fit this question, we take the top and bottom ESG score firms separated by regions (North America & Europe) to set up monthly portfolio return time series between 2002 and 2017 with either market value weighted or equally weighted returns.

As we control for risk factors based on research by Fama & French (1993), we find that using the VIX as a proxy for market volatility in the North American data set and a volatility index based on the STOXX50 as a proxy for market volatility in the European data set yields conflicting results. Multiple regression analysis reveals that North American ESG portfolios with only top score firms have a positive relation to volatility, while European ESG portfolios with only top score firms have a negative relation to volatility. Additionally, we find no plausible explanatory power of volatility with regard to portfolios with only bottom score

(3)

firms in either region, nor do we find relation in portfolios consisting of a long position in top score firms and a short position in bottom score firms.

Due to the conflicting results, we recommend for future researchers to either focus on the difference between top ESG score firms in North America and Europe or the constituents of the ESG score with regard to volatility. The existence of a key underlying factor, either firm-specific or ESG constituent firm-specific, that correlates strongly with volatility and causes these conflicting results seems plausible at the very least.

2. Literature

This section is where existing literature is used to develop a general understanding of previously explored relationships between portfolio performance, responsible investing and volatility. We first give a brief insight in to how return differences between portfolios should behave according to financial theories regarding asset selection. We then examine if research en masse supports traditional theory by referring to meta analyses. Then, research in market volatility and its relation to portfolio performance will be explored before finally forming hypotheses.

Asset selection, in our case selecting the best-in-class ESG firms to invest in, is traditionally seen as a conflicting strategy to fully diversifying your portfolio (Chegut et al., 2011). In the best interest of a regular investor, you should try to maximise your return based on a set appetite for risk. As such, when ESG firms their return does not differ from the market return, a less diversified ESG portfolio would underperform.

The effects of asset selection on performance in the realm of responsible investing are thoroughly explored by Kempf & Osthoff (2007). By creating portfolios that contain long positions in stocks with high socially responsible ratings and short positions in stocks with low socially responsible ratings they uncover abnormal returns of up to 8.7% annually. In layman terms, they buy high rated stocks and sell low rated stocks to turn a profit. Their results indicate that, adjusted for risk, the return of high and or low rated stocks have a different return than the market portfolio.

(4)

The characteristics of return through socially responsible investing are explored in more depth by Galema et al. (2008), where focus lies more on underlying factors that decide the eventual rating of a firm. They suggest that one of the reasons for ambiguity concerning the existence of outperformance by responsible investing is due to conflicting factors that make up the aggregate scores, specifically firms that score high on diversity and environmental issues contribute to outperformance.

In order to establish the relation between ESG and performance en masse we first consider the meta-analysis by Friede et al. (2015). By aggregating evidence of more than 2200 individual empirical studies they give a recent, but more importantly, a broad overview of the current state of research. Their primary motive is to tackle the prevailing ambiguity regarding ESG and performance in the field of finance, for instance the conflicting results between Hong & Kacperczyk (2007) and Kempf & Osthoff (2007).

By reviewing portfolios, vote-count studies and predecessors in meta-analysis they conclude that, albeit small, a positive correlation exists between ESG scores and performance. Further delving into regional results also shows them that, on average in developed markets, North American data sets have a higher share of positive findings than their European and Asian/Australian counterparts.

The meta-analysis of Revelli & Viviani (2014), gives additional insight to the results of Friede et. al (2015) while also reverting to the stage of ambiguity. They argue that, while a large group of researchers show positive findings concerning the relationship between ESG and performance, these are consistently studies that have created their own SRI portfolios, the majority of which exclude management fees. Studies that research existing SRI funds find ESG investing to be costly, resulting in a negative relationship. This is in line with traditional financial theory, where restricted investments imply less financial efficiency (Chegut et al., 2011).

This tells us that the definition of a research question is of vital importance to a clear interpretation of its answer. We mention portfolios rather than funds in our research question, where we imply the exclusion of management and transaction costs.

(5)

In 1996 the Chicago Board Options Exchange launched the VIX (now known as VXO), an index measuring the volatility of the Standard & Poor’s 100. Later in 2003, this index was continued by the VIX as we know today, which measures the volatility of the Standard & Poor’s 500 as the wider sample of firms gives a more accurate view on future volatility.2_{In a practical sense,} it is generally used by investors to understand the direction of a bear or bull market (Bongiovanni et al., 2016).

Quite a number of papers exist that research how portfolios behave based on market volatility indices such as the VIX. Copeland & Copeland (1999) for instance, test for a custom variable derived from the VIX to include a lagged effect. They find both “Value” and “Large-cap” portfolios to outperform their counterparts during high volatility. The opposite seems to be true as well, both “Growth” and “Small-cap” portfolios outperform their counterparts during low volatility.

From a different perspective, Chau (2012) tests historical stock prices for correlation with VIX to an exhaustible extent and finds an inverse relation between volatility and the stock market. This is in line with research by Eraker (2004), who finds a good fit between jumps in volatility stock prices through option prices and stock market data.

To re-iterate, through the aforementioned literature, we assume both volatility and ESG scores to have a well-established relation to portfolio performance. However, this leaves us in the dark regarding the pivotal issue of our research question. Does volatility contain any interaction effect with the returns of portfolios restricted by ESG scores?

This leads us to the following hypotheses:

H1a: Risk-adjusted net excess return of portfolios with top ESG score firms have a significant

relation to market volatility

H1b: Risk-adjusted net excess return of portfolios with bottom ESG score firms have a

significant relation to market volatility

H2: The difference between risk-adjusted net excess return of portfolios with top or bottom

ESG score firms has a significant relation to market volatility

To fill the gap in literature regarding the interaction effect of ESG rankings and volatility we create high and low ESG score portfolios and test to which extent their performance and the difference therein is explained by volatility in through multiple regression analysis.

(6)

3. Methodology

The analysis of volatility and its relation to ESG portfolio performance is conducted in twofold. In the first instance, portfolios that consist of either high or low ESG score firms are constructed from a North American set of historical returns and are tested on a North American market portfolio with North American Fama-French factors and the volatility of the S&P500 index. In the second instance of testing, portfolios of either high or low ESG score firms are constructed from a European set of historical returns and are tested on a European market portfolio with European Fama-French factors and the volatility of the STOXX50 index. The construction of our portfolios is similar to that of Kempf & Osthoff (2007) who test the difference between the top 10% rated firms and the bottom 10% rated firms. Which in turn is similar to the approach of Hong & Kacperczyk (2007) who created portfolios of sin stocks and comparables to test for a difference in return.

The dependent variables in this paper are the risk adjusted returns of portfolios consisting of only “Good” firms, only “Bad” firms or only the risk adjusted returns of portfolios consisting of long positions in “Good” firms and short positions in “Bad” firms. “Good” and “Bad” are labels that are placed on firms in the data set on a monthly frequency based on their relative ESG rating in the data set. Cut-off points for “Good” firms are at the top 2.5%, 5% and 10%. Cut-off points for “Bad” firms are set at the bottom respectively.

The weights of firms in the portfolios are rebalanced every month to account for the event of a firm their score to cross the portfolio threshold. However, the ESG score of a firm is not the only measure that decides the inclusion and weight in the eventual portfolio. As is conventional, we will weigh the individual firm their return to their market value relative to the rest of the firms their combined market value in the respective portfolio.

With the ESG scores and market values in place, we can use them in conjunction with the total return to construct the time series of the “Good” and “Bad” portfolio return. The monthly return of firms will calculated as follows for the “Good” and “Bad” portfolios:

𝑅_𝑖,𝑡 = ∑𝐾𝑘=1 𝑀𝑉𝑘,𝑡_{𝑀𝑉𝑖,𝑡} ×𝑅𝑘,𝑡− 𝑅𝑓𝑡 (1)

(7)

Subsequently, for the long “Good” short “Bad” portfolio:

𝑅𝑔𝑚𝑏𝑖,𝑡 = 𝑅𝑔𝑖,𝑡 − 𝑅𝑏𝑖,𝑡 (2)

where i is the portfolio at the 2.5%, 5% or 10% level, 𝑅𝑔𝑚𝑏𝑖,𝑡 is the return of the “Good Minus Bad” portfolio i at month t, 𝑅𝑔𝑖,𝑡 is the return of the “Good” portfolio i at month t and 𝑅𝑏𝑖,𝑡 is the return of the “Bad” portfolio i at month t.

Calculations 1 and 2 are performed in both the North American and European data set with ESG criteria to create “Good”, “Bad” and “Good Minus Bad” portfolios at the 2.5%, 5% and 10% level.

As control variables we use the common risk factors from Fama and French (1993). The market risk premium (MRP) captures any relation of the portfolios to common market variation. The size factor (SMB) controls for portfolio characteristics such as a predominantly high market value of constituent firms. The book-to-market factor (HML) controls for portfolios that are over-weighted with either value or growth firms.

We forego controlling for the momentum factor which is commonly used in conjunction with the size factor and book-to-market factor, as we expect an existing relation between momentum and volatility that might cause excessive noise in regression analysis. Research has been done on this relation, where Wang and Xu (2014) find a predictive power in volatility with regard to momentum pay-offs and Arena et. al. (2008) find a significant relation between market volatility and momentum portfolio returns.

For the uncertainty effect that we try to include, we are using the volatility indices of the S&P500 and the STOXX50 as proxies for North American and European markets respectively. These indices show the percentages that markets are expected to rise or fall in the next 30 days. An alternative to this are Bollinger Bands, which denote the space between two lines that widens during high volatility and lessens during low volatility. In the absence of a clear-cut difference from volatility indices, we opt not to use Bollinger Bands.

(8)

𝑅_𝑡 = 𝛼 + 𝛽₁𝑀𝑅𝑃_𝑡+ 𝛽₂𝑆𝑀𝐵_𝑡 + 𝛽₃𝐻𝑀𝐿_𝑡+ 𝜀_𝑡 (3)

𝑅𝑡 = 𝛼 + 𝛽1𝑀𝑅𝑃𝑡+ 𝛽2𝑆𝑀𝐵𝑡 + 𝛽3𝐻𝑀𝐿𝑡+ 𝛽4𝑉𝐼𝑋𝑡+ 𝜀𝑡 (4)

where 𝑅𝑡 is the return of the “Good”, “Bad” or “Good Minus Bad” portfolio at month t, 𝛼 is the constant common intercept, 𝑀𝑅𝑃𝑡 is the market risk premium at month t, 𝑆𝑀𝐵𝑡 is size factor at month t, 𝐻𝑀𝐿𝑡 is the book-to-market factor at month t, 𝑉𝐼𝑋𝑡 is volatility index at month t and 𝜀𝑡 is the error term at month t.

In order to confirm the hypotheses, we need to look at the coefficient of VIX and its significance in the regression analysis of each portfolio. As volatility is sparingly used as an indicator of financial markets, it is not improbable for it to have a significant impact on the return of the “Good” and “Bad” portfolios.

4. Data analysis

This section serves two major purposes. One, to provide an in-depth, intensive, view on both data sets. Two, to reconfigure the data sets into several tabular formats to give an extensive bird’s-eye view. The combination serves to give a deeper understanding of the underlying data of the portfolios as well as the independent variables.

We first explore the data components of the portfolio return time series, namely the firms their ESG scores, market values and returns. This is followed by a view on the descriptive statistics of the subsequent portfolio returns. Lastly, we analyse the independent variables: risk factors and volatility indices.

(9)

Table 1: Descriptive Statistics for the Environmental, Social and Governance Score of North American and European firms between 2002 and 2017

This table presents the descriptive statistics (mean, minimum, maximum, median, standard deviation and observations) of firms their Environmental, Social and Governance Score between 2002 and 2017. Descriptive Statistics are aggregated per year; underlying data is on a monthly basis. Panel A shows scores for the North American data set, Panel B shows scores for the European data set. Source: DataStream.

Panel A: North America

Year Mean Min Max Median Std. Dev N

2002 47.07 0.00 85.27 44.75 14.02 3792 2003 47.42 0.00 87.33 45.88 14.02 4008 2004 48.29 0.00 91.94 46.99 13.40 6263 2005 48.91 0.00 92.83 46.96 14.80 6924 2006 49.46 0.00 95.96 46.73 15.40 7104 2007 48.73 0.00 95.54 46.83 16.13 8308 2008 47.94 0.00 96.71 44.85 17.20 10874 2009 48.67 0.00 97.63 46.06 17.63 11963 2010 49.79 0.00 97.63 47.48 17.37 12691 2011 49.79 0.00 93.93 48.12 17.69 13088 2012 49.78 0.00 91.75 48.93 17.38 13245 2013 49.97 0.00 91.93 49.28 17.34 13658 2014 49.82 0.00 94.23 48.59 17.07 14592 2015 47.41 11.61 94.23 45.15 16.69 22958 2016 46.34 8.08 93.65 43.56 16.19 31603 2017 44.61 11.46 95.35 41.98 14.93 26521 Total 47.97 0.00 97.63 45.59 16.50 207592 Panel B: Europe

(10)

There are two important points of evaluation that follow the descriptive statistics of Table 1. Observations steadily grow in volume up to 2016 in both data sets, any results that are later inferred on the basis of the “Good” and “Bad” portfolios are not equally representative of the 2002-2017 period, but predominantly the second half of the time period.

Furthermore, firms with an ESG rating of 0.00 appear in the statistics above. These are not firms that did not get a score, but simply received the lowest score. As such, they are not excluded from the “Bad” portfolios.

The descriptive statistics of Table 2 show us some clear differences between subsets. First and foremost, “Good” firms are larger than “Bad” firms in both the North American and European data sets. Any effect this has on either the return of the eventual portfolios and the long/short portfolios is controlled for by the “Small Minus Big” factor as constructed by Fama & French (1993).

Table 2: Descriptive Statistics for the Market Value of North American and European subsets between 2002 and 2017

This table presents the descriptive statistics (mean, minimum, maximum, median, standard deviation and observations) of firms their Market Value in millions of dollars between 2002 and 2017. Descriptive Statistics are aggregated per portfolio; underlying data is on a monthly basis per firm in the subset. Panel A shows Market Value for the North American data set, Panel B shows Market Value for the European data set. Source: DataStream.

Mean Min Max Median Std. Dev N

Good (2.5%) 66757.88 577.76 847096.70 34664.67 82047.39 5243 Good (5%) 60133.19 288.57 847096.70 30791.76 77180.72 10355 Good (10%) 49499.47 205.72 847096.70 23397.21 69231.81 20492 Bad (10%) 5188.55 0.85 272741.40 2359.85 14167.45 19583 Bad (5%) 4537.94 0.94 261716.70 2239.21 12117.94 9697 Bad (2.5%) 3628.61 0.94 89403.25 2150.07 5558.58 4767 Panel B: Europe

Good (2.5%) 38009.94 482.81 355807.50 30726.32 32953.98 3765 Good (5%) 38985.65 168.41 845663.40 26647.22 51559.73 7363 Good (10%) 39698.91 61.02 845663.40 19354.38 66353.46 14633 Bad (10%) 6327.99 2.68 496887.10 1947.79 20717.28 14153 Bad (5%) 5154.92 2.68 132471.50 2083.99 9686.96 7103 Bad (2.5%) 4456.32 2.68 132471.50 1992.79 7927.16 3532

(11)

a significant margin, revealing that a substantial number of smaller firms join the portfolio. The European “Good” portfolio shows a different change in composition as the cut-off point moves from the top 5% to the top 10%. Its mean does not shift by a meaningful amount while the median does drop substantially. This seems to be an extension from the cut-off point change between 2.5% and 5% where larger firms join the portfolio. This is not reflected in the any of the “Bad” portfolios, wherein the mean gradually lowers as the cut-off point increasingly restricts the ESG score.

The eagle-eyed observer will also see that the subsets their statistics pertaining to market value have less observations than their ESG counterparts. This is due to market value statistics excluding zero value data points. The market value data occasionally ends interim year, while ESG scores set at the start of every year. This does not affect the constructed portfolios as a firm does have a return when there is no reported market value, regardless of their ESG score. With data sets of monthly ESG scores established, we move forward with the construction of “Good” and “Bad” portfolios. Table 3 shows the descriptive statistics of ESG Scores of the constructed portfolios their constituent firms.

Table 3: Descriptive Statistics for the Environmental, Social and Governance Score of North American and European subsets between 2002 and 2017

This table presents the descriptive statistics (mean, minimum, maximum, median, standard deviation and observations) of firms their Environmental, Social and Governance Score between 2002 and 2017. Descriptive Statistics are aggregated per portfolio; underlying data is on a monthly basis per firm in the subset. Panel A shows scores for the North American data set, Panel B shows scores for the European data set. Source: DataStream.

Total 47.97 0.00 97.63 45.59 16.50 207592 Good (2.5%) 85.67 77.57 97.63 85.25 3.33 5277 Good (5%) 82.62 71.05 97.63 82.17 4.25 10453 Good (10%) 78.44 65.77 97.63 77.93 5.64 20817 Bad (10%) 23.55 0.00 33.21 24.34 4.68 20806 Bad (5%) 20.61 0.00 30.86 21.53 4.73 10456 Bad (2.5%) 17.91 0.00 27.79 19.18 5.07 5282 Panel B: Europe

(12)

Table 3 shows that on average European firms beat North American firms by about ten percentage points in ESG scores, however this difference is less pronounced in the “Good” and “Bad” portfolios. The lowest observation count is shown to be 3765, that of the European “Good” 2.5%. This translates to, on average, the top and bottom portfolios consisting of 20 firms. Whilst this is on the low end of observations, the timespan covers 192 months and, additionally, the portfolios get rebalanced every month which increases the diversity of firms in the portfolios. As such, the relatively low minimum of firms in portfolios does not impair us in our process of analysing portfolios. More detailed descriptive statistics of these subsets are included in Appendix B.

With all data underlying our portfolios analysed, we move to the descriptive statistics of our dependent variable in Table 4.

Most notably in Table 4, the “GMB” (Good Minus Bad) portfolios seem to show a consistent negative mean across regions and ESG criteria. This means that “Bad” portfolios, on average, consistently outperform “Good” portfolios. The return statistics of both regions across cut-off points are roughly identical across mean, median and standard deviation. This allows for these regions their portfolios to undergo independent analyses and as such, act as robustness checks for one another.

As an additional important note, the return time series of these portfolios are adjusted for risk. The risk-free return data for both North-American and European data sets is retrieved from Kenneth R. French his data library3_.

Table 4: Descriptive Statistics for the Total Return of North American and European portfolios between 2002 and 2017

This table presents the descriptive statistics (mean, minimum, maximum, median, standard deviation and observations) of monthly returns between 2002 and 2017. “Good”, “Bad” and “GMB” denote market value weighted portfolios of the top x% ESG score, bottom x% ESG score or the long top x% short bottom x% ESG score firms respectively. Descriptive Statistics are aggregated per portfolio; underlying data is on a monthly basis. Panel A shows portfolios for the North American data set, Panel B shows portfolios for the European data set. Source: DataStream.

(13)

Good (2.5%) 0.95 -16.52 14.87 1.46 4.66 192 Bad (2.5%) 1.80 -17.48 15.83 2.03 5.07 192 GMB (2.5%) -0.85 -11.52 15.08 -1.07 3.63 192 Good (5%) 0.91 -14.41 14.90 1.14 4.04 192 Bad (5%) 1.61 -18.36 20.28 2.10 5.02 192 GMB (5%) -0.70 -11.47 11.18 -0.92 3.35 192 Good (10%) 0.91 -15.02 11.57 1.43 3.88 192 Bad (10%) 1.44 -16.23 16.56 1.68 4.58 192 GMB (10%) -0.53 -8.17 7.15 -0.60 2.41 192 Panel B: Europe

Good (2.5%) 0.80 -17.73 16.78 1.53 5.42 192 Bad (2.5%) 1.62 -23.78 23.66 2.14 6.00 192 GMB (2.5%) -0.82 -10.98 11.17 -0.80 3.99 192 Good (5%) 0.79 -18.27 15.30 1.20 5.15 192 Bad (5%) 1.08 -20.32 19.99 1.49 5.26 192 GMB (5%) -0.29 -8.62 9.33 -0.36 3.30 192 Good (10%) 0.80 -13.70 17.21 1.21 4.66 192 Bad (10%) 1.15 -17.86 21.58 1.44 4.62 192 GMB (10%) -0.36 -9.20 10.67 -0.45 2.67 192

In Table 5 we show the descriptive statistics of our control variables and the volatility indices. To illustrate the practical implications of these numbers, the maximum of the volatility index in North America (VIX) is 59.89. At this point in time the market expected share prices to either rise or fall by roughly 60 percent by next month, or 30 days to be exact.

The SMB and HML variables represent the percentage return of the long/short portfolios in our time range, but as initially constructed by Fama and French (1993). The MRP is interpreted as the monthly percentage return reward for a unit of risk.

(14)

Table 5: Descriptive Statistics for the independent variables of North American and European portfolios between 2002 and 2017

This table presents the descriptive statistics (mean, minimum, maximum, median, standard deviation and observations) of the market risk premium (MRP), the size factor (SMB), the book-to-market factor (HML) and volatility indices (VIX) between 2002 and 2017. Underlying data is on a monthly basis. Panel A shows statistics for the North American data set, Panel B shows statistics for the European data set. Source: DataStream, Kenneth R. French.

MRP 0.64 -17.23 11.35 1.09 4.12 192

SMB 0.22 -5.30 6.11 0.17 2.37 192

HML 0.09 -11.10 8.27 -0.10 2.48 192

VIX 19.30 9.51 59.89 16.80 8.37 192

Panel B: Europe

MRP 0.68 -22.02 13.67 0.77 5.31 192

SMB 0.25 -6.83 4.88 0.29 1.90 192

HML 0.24 -4.35 7.42 0.29 2.16 192

VIX 23.97 11.99 61.34 21.69 9.35 192

5. Results

In this section we explore the various results of multiple regression analyses. We combine the ‘Capital Asset Pricing Model’ (Sharpe, 1964) with the ‘Fama-French three-factor model (Fama & French, 1993). After the individual portfolios are examined, we add the volatility indices to the model. Finally, these steps are reproduced for the long/short portfolio.

With such a wide range of results to interpret in Table 6.1 alone, we start at a single measure for the individual portfolios and widen the scope from there. Concerning the top 5% ESG portfolio of North America (“Good” 5%), we see an intercept of 0.35 significant at the 1% level. Moving the ESG cut-off point to 2.5% and 10% it seems to retain its significance but widening the range of the intercept to 0.31 and 0.39. This is reflected in the European data set at 0.40. This shows that there is an alpha, or rather net gain, to be made by investing in high ESG score firms of about 0.35% on a monthly basis. However, moving over to the bottom ESG score firms, this measure retains its significance but can be anywhere from 0.52% to 1.11% depending on the portfolio.

(15)

will net a loss of anywhere between 0.44% to 0.80% on a monthly basis. As the restriction on ESG scores become less stringent, the difference decreases but it also loses some of its significance at less stringent cut-off points.

In layman terms, you could say that there is a price to be paid for restricting your portfolio to high ESG score firms compared to bottom ESG score firms in North America. Depending on the cut-off point in ESG scores, the difference between portfolios can be anywhere from a 3% annual return, up to 6.5%. This seems to be in line with “The Price of Taste” by Dam et. al. (2017), where they estimate the price of SRI investing at 4.8% annually.

Table 6.1: The effect of common risk factors and volatility indices on monthly returns of market value weighted North American portfolios between 2002 and 2017

This table presents the coefficients, t-statistics and significance of common risk factors, the market risk premium (MRP), the size factor (SMB), the book-to-market factor (HML) and additionally the volatility indices (VIX) of monthly returns between 2002 and 2017. “Good”, “Bad” and “Good-minus-bad” denote market value weighted portfolios of the top x% ESG score, bottom x% ESG score or the long top x% short bottom x% ESG score firms respectively. Model 1 and 2 show the results of multivariate analysis with the following equations respectively: Rt= α + β1 MRPt+ β2 SMBt + β3 HMLt+ εt, Rt= α + β1 MRPt+ β2

SMBt + β3 HMLt+ β4 VIXt + εt. The regressions are estimated with 192 observations. Numbers in brackets

are t-statistics. *, **, *** indicate significance at the 0.10, 0.05 and 0.01 level respectively. Source: DataStream. Kenneth R. French.

Panel A: North America at a 2.5% ESG criterium

Good (2.5%) Bad (2.5%) Good-minus-bad (2.5%)

(16)

Panel B: North America at a 5% ESG criterium

Good (5%) Bad (5%) Good-minus-bad (5%)

Model (1) (2) (1) (2) (1) (2) Intercept 0.35*** -0.34 0.92*** 1.66*** -0.57** -2.00*** [3.44] [-1.20] [4.80] [3.07] [-2.48] [-3.13] MRP 0.95*** 0.99*** 1.01*** 0.97*** -0.05 0.01 [36.3] [34.53] [20.48] [18.02] [-0.91] [0.17] SMB -0.29*** -0.30*** 0.22** 0.23*** -0.51*** -0.53*** [-6.34] [-6.63] [2.56] [2.68] [-4.99] [-5.24] HML 0.03 0.05 -0.12 -0.13* 0.15 0.18* [0.73] [1.10] [-1.47] [-1.66] [1.55] [1.90] VIX 0.04** -0.04 0.07** [2.60] [-1.46] [2.4] R Square 0.88 0.89 0.73 0.73 0.14 0.17

Panel C: North America at a 10% ESG criterium

Model (1) (2) (1) (2) (1) (2) Intercept 0.35*** -0.18 0.79*** 0.44 -0.44*** -0.62 [4.79] [-0.87] [5.31] [1.05] [-2.65] [-1.31] MRP 0.93*** 0.96*** 1.00*** 1.02*** -0.07 -0.06 [49.61] [46.97] [26.16] [24.1] [-1.64] [-1.32] SMB -0.22*** -0.23*** 0.04 0.04 -0.27*** -0.27*** [-6.82] [-7.15] [0.68] [0.60] [-3.60] [-3.62] HML 0.07** 0.08*** -0.13** -0.12* 0.19*** 0.20*** [2.23] [2.64] [-2.05] [-1.91] [2.82] [2.84] VIX 0.03*** 0.02 0.01 [2.77] [0.88] [0.40] R Square 0.93 0.94 0.80 0.81 0.11 0.12

(17)

portfolio, which we confirm through the descriptive statistics in Table 6. The “Good” portfolios have even lower coefficients between -0.38 and -0.62, significant at the 1% level. Which in the same way denotes a primarily large cap portfolio, in line with the aforementioned descriptive statistics. The “Bad” portfolios do not consistently have significant results on this measure, but those that do are positive – indicating characteristics of a small cap portfolio. The long/short portfolios consistently have a negative coefficient significant at the 1% level, indicating that the small cap characteristics of the short “Bad” portfolios do not seem to outweigh the effects of the large cap characteristics of the long “Good” portfolios.

As a final note on the control variables, the High Minus Low (HML) coefficient only shows a significance in a few portfolios across regions, most notably at the 10% portfolios of the North American data set. The “Good” 10% North American portfolio has a positive 0.07 coefficient at the 5% significance level, the “Bad” 10% portfolio has a negative -0.13 coefficient at the same significance. This indicates that the “Good” portfolio is leaning towards the characteristics of a “Value” fund, while the “Bad” portfolio is leaning towards a “Growth” fund. The other portfolios with significant HML loadings show similar signs, as perpetuated by the ten percent “Good-minus-bad” North American HML coefficient of 0.19. While not consistent across regions and cut-off points, this does indicate that high ESG score portfolios might be “Value” oriented.

Table 6.2: The effect of common risk factors and volatility indices on monthly returns of market value weighted European portfolios between 2002 and 2017

This table presents the coefficients, t-statistics and significance of common risk factors, the market risk premium (MRP), the size factor (SMB), the book-to-market factor (HML) and additionally the volatility indices (VIX) of monthly returns between 2002 and 2017. “Good”, “Bad” and “Good-minus-bad” denote market value weighted portfolios of the top x% ESG score, bottom x% ESG score or the long top x% short bottom x% ESG score firms respectively. Model 1 and 2 show the results of multivariate analysis with the following equations respectively: Rt= α + β1 MRPt+ β2 SMBt + β3 HMLt+ εt, Rt= α + β1 MRPt+ β2

SMBt + β3 HMLt+ β4 VIXt + εt. The regressions are estimated with 192 observations. Numbers in brackets

(18)

Panel A: Europe at a 2.5% ESG criterium

Model (1) (2) (1) (2) (1) (2) Intercept 0.40* 2.10*** 0.97*** 3.14*** -0.57** -1.04 [1.83] [3.26] [3.38] [3.70] [-2.09] [-1.27] MRP 0.78*** 0.73*** 0.81*** 0.75*** -0.03 -0.02 [17.11] [15.08] [13.55] [11.74] [-0.59] [-0.32] SMB -0.62*** -0.64*** 0.16 0.13 -0.78*** -0.77*** [-5.43] [-5.70] [1.05] [0.88] [-5.45] [-5.38] HML 0.13 0.14 0.25* 0.26* -0.12 -0.12 [1.18] [1.25] [1.68] [1.77] [-0.83] [-0.84] VIX -0.07*** -0.09*** 0.02 [-2.8] [-2.71] [0.61] R Square 0.70 0.72 0.58 0.60 0.14 0.15

Panel B: Europe at a 5% ESG criterium

Model (1) (2) (1) (2) (1) (2) Intercept 0.40* 1.41** 0.52** 2.53*** -0.12 -1.12 [1.89] [2.24] [2.10] [3.49] [-0.53] [-1.65] MRP 0.76*** 0.73*** 0.74*** 0.68*** 0.02 0.05 [17.26] [15.43] [14.37] [12.47] [0.41] [0.97] SMB -0.52*** -0.54*** 0.10 0.08 -0.62*** -0.61*** [-4.75] [-4.88] [0.77] [0.59] [-5.25] [-5.15] HML 0.06 0.06 0.17 0.18 -0.11 -0.11 [0.54] [0.58] [1.34] [1.43] [-0.95] [-0.99] VIX -0.04* -0.08*** 0.04 [-1.71] [-2.95] [1.56] R Square 0.70 0.70 0.60 0.62 0.14 0.15

Panel C: Europe at a 10% ESG criterium

(19)

Looking at Table 6.2 next to Table 6.1, the signs and coefficients of the control variables in the European data do not differ enough from the North American data set as to invalidate one or the other. The notable differences appear in the significant relation of the alpha of the long/short portfolios. While this is significant on all cut-off points in the North American data set, the European portfolios lose a significant alpha as the cut-off point becomes less stringent. The significant alpha of Table 6.2, Panel C, “Good-minus-bad 10%” might indicate that there is too much noise in Panel C without the inclusion of the volatility factor.

The volatility indices, as added in equation (2), show conflicting signs across regions in the individual portfolios. In the North American data set, all “Good” portfolios show coefficients at significant levels, these coefficients range between 0.03 and 0.04. While the “Bad” portfolios do not have any significant relation to the volatility index, the long/short portfolio does have one at the 2.5 and ten percent cut-off points. The coefficients range between 0.06 and 0.07, showing that in the North American data set the high ESG score portfolios benefit from a higher volatility, and even more so when compared to bottom ESG score portfolios. The benefit seems to be that for every one unit increase of the volatility index, the return of a portfolio increases by roughly 0.06% ceteris paribus.

The opposite effect seems to be at play in the European data set. All “Good” and “Bad” portfolios have coefficients at significant levels, all of them negative. Across all “Good” and “Bad” portfolios, the effect of a one unit increase of the volatility index, the return of a portfolio decreases by anywhere between -0.03% and -0.09% ceteris paribus. At face value there seem to be minor differences between the volatility index coefficients of “Good” and “Bad” portfolios, but the long/short “Good-minus-bad” portfolios show no significant coefficients.

Considering the vastly different signs between the North American and European portfolios regarding their respective volatility indices, we should first set up a second approach before assuming a regional difference might be occurring. Hence, as both a second means of testing for results and a robustness check, we repeat the exact same steps of analysis but use equally weighted portfolios rather than market value weighted portfolios.

(20)

Table 7.1: The effect of common risk factors and volatility indices on monthly returns of equally weighted North American portfolios between 2002 and 2017

This table presents the coefficients, t-statistics and significance of common risk factors, the market risk premium (MRP), the size factor (SMB), the book-to-market factor (HML) and additionally the volatility indices (VIX) of monthly returns between 2002 and 2017. “Good”, “Bad” and “Good-minus-bad” denote equally weighted portfolios of the top x% ESG score, bottom x% ESG score or the long top x% short bottom x% ESG score firms respectively. Model 1 and 2 show the results of multivariate analysis with the following equations respectively: Rt= α + β1 MRPt+ β2 SMBt + β3 HMLt+ εt, Rt= α + β1 MRPt+ β2 SMBt

+ β3 HMLt+ β4 VIXt + εt. The regressions are estimated with 192 observations. Numbers in brackets are

t-statistics. *, **, *** indicate significance at the 0.10, 0.05 and 0.01 level respectively. Source: DataStream. Kenneth R. French.

Panel A: North America at a 2.5% ESG criterium

Model (1) (2) (1) (2) (1) (2) Intercept 0.19* -0.01 0.58*** 0.32 -0.39** -0.33 [1.85] [-0.04] [3.44] [0.67] [-2.20] [-0.65] MRP 1.02*** 1.03*** 0.98*** 0.99*** 0.05 0.05 [39.31] [35.93] [22.53] [20.65] [1.05] [0.90] SMB -0.03 -0.03 0.39*** 0.39*** -0.42*** -0.42*** [-0.60] [-0.66] [5.27] [5.19] [-5.33] [-5.29] HML 0.12*** 0.13*** 0.15** 0.15** -0.02 -0.02 [2.98] [3.05] [2.09] [2.15] [-0.29] [-0.30] VIX 0.01 0.01 0.00 [0.74] [0.58] [-0.13] R Square _0.91 _0.91 _0.79 _0.79 _0.14 _0.14

Panel B: North America at a 5% ESG criterium

(21)

Panel C: North America at a 10% ESG criterium

Model (1) (2) (1) (2) (1) (2) Intercept 0.21*** -0.23 0.52*** 0.26 -0.31*** -0.49 [2.81] [-1.12] [4.71] [0.82] [-2.93] [-1.63] MRP 1.00*** 1.02*** 1.02*** 1.04*** -0.02 -0.02 [52.01] [48.66] [36.01] [33.04] [-0.90] [-0.55] SMB 0.02 0.01 0.36*** 0.36*** -0.35*** -0.35*** [0.49] [0.30] [7.42] [7.31] [-7.38] [-7.4] HML 0.11*** 0.12*** 0.11** 0.12*** 0.00 0.00 [3.64] [3.96] [2.52] [2.62] [-0.06] [0.03] VIX 0.02** 0.01 0.01 [2.26] [0.90] [0.64] R Square 0.94 0.95 0.90 0.91 0.26 0.26

Table 7.2: The effect of common risk factors and volatility indices on monthly returns of equally weighted European portfolios between 2002 and 2017

This table presents the coefficients, t-statistics and significance of common risk factors, the market risk premium (MRP), the size factor (SMB), the book-to-market factor (HML) and additionally the volatility indices (VIX) of monthly returns between 2002 and 2017. “Good”, “Bad” and “Good-minus-bad” denote equally weighted portfolios of the top x% ESG score, bottom x% ESG score or the long top x% short bottom x% ESG score firms respectively. Model 1 and 2 show the results of multivariate analysis with the following equations respectively: Rt= α + β1 MRPt+ β2 SMBt + β3 HMLt+ εt, Rt= α + β1 MRPt+ β2 SMBt

+ β3 HMLt+ β4 VIXt + εt. The regressions are estimated with 192 observations. Numbers in brackets are

t-statistics. *, **, *** indicate significance at the 0.10, 0.05 and 0.01 level respectively. Source: DataStream. Kenneth R. French.

Panel A: Europe at a 2.5% ESG criterium

(22)

Panel B: Europe at a 5% ESG criterium

Model (1) (2) (1) (2) (1) (2) Intercept 0.12 1.52** 0.30 1.54** -0.18 -0.03 [0.56] [2.39] [1.47] [2.54] [-1.19] [-0.06] MRP 0.78*** 0.74*** 0.73*** 0.70*** 0.05 0.04 [17.51] [15.53] [17.24] [15.31] [1.55] [1.29] SMB -0.42*** -0.43*** 0.34*** 0.32*** -0.75*** -0.76*** [-3.73] [-3.92] [3.15] [3.03] [-9.54] [-9.52] HML 0.27** 0.28** 0.27*** 0.28*** 0.00 0.00 [2.50] [2.57] [2.63] [2.70] [-0.01] [-0.01] VIX -0.06** -0.05** -0.01 [-2.33] [-2.17] [-0.36] R Square 0.72 0.72 0.70 0.70 0.35 0.35

Panel C: Europe at a 10% ESG criterium

Model (1) (2) (1) (2) (1) (2) Intercept 0.15 1.47*** 0.27 1.13** -0.12 0.33 [0.83] [2.76] [1.51] [2.11] [-1.09] [1.00] MRP 0.77*** 0.73*** 0.72*** 0.70*** 0.05* 0.03 [20.41] [18.21] [19.20] [17.23] [1.93] [1.25] SMB -0.30*** -0.32*** 0.44*** 0.43*** -0.74*** -0.75*** [-3.18] [-3.40] [4.66] [4.56] [-12.61] [-12.71] HML 0.22** 0.23** 0.32*** 0.32*** -0.10* -0.09* [2.44] [2.53] [3.49] [3.54] [-1.69] [-1.67] VIX -0.05*** -0.04* -0.02 [-2.63] [-1.70] [-1.45] R Square 0.77 0.77 0.75 0.75 0.48 0.49

Several changes in signs and significance of coefficients occur by switching to equally weighted portfolios. Rather remarkably, the fit of the models (R Square) increases by a noticeable margin. From the perspective of the Fama & French (1993) factors in regression 1 (Table 7.1 and 7.2), the equally weighted portfolios should be a better representation of the market if the factors fit better than they do with market weighted portfolios.

(23)

American data set, the signs and significance of the alphas are similar, but they are smaller across all three portfolios and all cut-off points. Essentially, the established price of ESG investing still stands after a robustness check but is more likely to be on the lower end of the aforementioned range. Roughly at a cost between 2.5% to 3.5% in annual return.

The Market Risk Premium (MRP) is still at a highly significant level for both “Good” and “Bad” portfolios across regions and has more closely converged to 1 compared to the previous test. Oddly enough, it has picked up a coefficient of 0.05 for the European long/short portfolio at a cut-off point of 10%, significant at the 10% level. With no apparent theoretical explanation at hand and it being a single occurrence, we choose to ignore this and attribute it to noisy data.

The Small Minus Big (SMB) coefficients lose and gain significance in both regions compared to the market value weighted portfolios, but the signs do not change. Furthermore, the coefficients do not lose meaning when compared to the descriptive statistics of market values in Table 2. As such, we expect the high average market value “Good” portfolios to have a negative sign and the low average market value “Bad” portfolios to have a positive sign. The High Minus Low (HML) coefficient has noticeably gained significance, with positive signs for both “Good” and “Bad” portfolios in both regions. Interestingly, the long/short portfolio in the European data set at the 10% cut-off point has a coefficient of -0.10 at a 10% significance level, an opposite sign to the only long/short portfolio in the market value weighted (North American) significant result for the HML variable. The distribution of “Value” and “Growth” firms seems to have changed after weighing firms equally in our portfolios. The equally weighted portfolios show the same signs for the volatility index as the market value weighted portfolios do, indicating a positive relation in the North American data set, but indicating a negative relation in the European data set. Thus, the robustness check with equally weighted portfolios gives no new insights in a relation between volatility and the difference between “Good” and “Bad” portfolios.

As for the first hypothesis (H1a), across both equally weighted and market value weighted

tests, four out of six North American “Good” portfolios and six out of six European “Good” portfolios indicate a significant relation to their respective volatility indices. This gives plausible gravitas to confirm the hypothesis of a relation between the measures. With regard to the second hypothesis (H1b), none of the North American “Bad” portfolios and five out of

(24)

between high and low ESG score firms. With only two significant volatility coefficients in the market value weighted North American data set and only one significant volatility coefficient in the equally weighted European data set, no consistent proof across regions and data sets exists.

6. Concluding remarks

During the inception of and initial background reading for this paper, expectations of certain effects of the independent variables on the portfolio returns were quickly raised. Most notably how the control variables would give insights into the portfolios that were created from a purely data-oriented approach with arbitrary restrictions.

Including a proxy for volatility or rather, uncertainty, in a long/short restricted ESG portfolio was decidedly a dive into the deep end as there is not much prior research into this specific combination of variables. Logically and anecdotally, you can string together reasonably valid expectations for a relation, but whether it turns out to be backed by data and also economically significant is another matter altogether.

The implications of the results explained in the last segment are quite clear, in the case of “Good” portfolios in Europe there is a relation to the volatility index of the STOXX50, with an opposite sign to the similar relation that “Good” portfolios have in North America to the volatility index of the S&P500. Positive findings for the remaining hypotheses are few and far between.

In a practical sense, the results give individual investors reason to under-weight high ESG score European firms in their portfolio during periods of high volatility. Conversely, to over-weight highly ESG score North American firms during periods of high volatility. The big question that arises from the above statements is why these relations to volatility conflict between regions.

(25)

even 178 minor constituent scores (Appendix A) and pinpoint which score relates most to the movement of volatility indices.

(26)

References

Arena, Matteo P., Haggard, K. Stephen and Yan, Xuemin, Sterling, 2015, Price Momentum and Idiosyncratic Volatility, The Financial Review 43, 159-190.

Bernstein, William J., Rolling Your Own: Three-Factor Analysis, Efficient Frontier,

http://www.efficientfrontier.com/ef/101/roll101.htm.

Bongiovanni, Alessio, Paola De Vicentiis, and Eleonora Isaia, 2016, Performance of VIX index as a tool for volatility forecasting in bear and bull markets, 9th Annual conference of

EuroMed Academy of Business.

Chau, Hy, 2012, The VIX and its applications in the market, The University of Utah

Chegut, Andrea, Hans Schenk, and Bert Scholtens, 2011, Assessing SRI fund performance research: Best practices in empirical analysis, Sustainable Development 19, 77–94.

Copeland, Maggie M., and Thomas E. Copeland, 1999, Market Timing: Style and Size Rotation Using the VIX, Financial Analysts Journal 55, 73–81.

Ciciretti, Rocco, Dalò, Ambrogio and Dam, Lammertjan, 2008, The Price of Taste for Socially Responsible Investment. (Working Paper)

Dam, Lammertjan, and Bert Scholtens, 2015, Toward a theory of responsible investing: On the economic foundations of corporate social responsibility, Resource and Energy Economics 41, 103–121.

Desclée, Albert, Jay Hyman, Lev Dynkin, and Simon Polbennikov, 2016, Sustainable investing and bond returns - Investment Bank, Barclays.

(27)

Fama, Eugene F., and Kenneth R. French, 1993, Common risk factors in the returns on stocks and bonds, Journal of Financial Economics 33, 3–56.

Friede, Gunnar, Timo Busch, and Alexander Bassen, 2015, ESG and financial performance: aggregated evidence from more than 2000 empirical studies, Journal of Sustainable Finance & Investment 5, 210–233.

Galema, Rients, Plantinga, Auke and Scholtens, Bert, 2008, The stocks at stake: Return and risk in socially responsible investment, Journal of Banking & Finance 32, 2646-2654.

Kacperczyk, Marcin T., and Harrison G. Hong, 2006, The Price of Sin: The Effects of Social Norms on Markets, Journal of Financial Economics 93, 15-36.

Kempf, Alexander and Osthoff, Peer, 2007, The Effect of Socially Responsible Investing on Portfolio Performance, European Financial Management 13, 908-922.

Nagy, Zoltán, Altaf Kassam, and Linda-Eling Lee, 2016, Can ESG Add Alpha? An Analysis of ESG Tilt and Momentum Strategies, The Journal of Investing 25, 113–124.

Revelli, Christophe, and Jean-Laurent Viviani, 2014, Financial performance of socially responsible investing (SRI): what have we learned? A meta-analysis, Business Ethics: A European Review 24, 158–185.

Schaeffer, VIX and VXO, Schaeffers Research,

https://www.schaeffersresearch.com/education/volatility-basics/volatility-indexes-etns/vix-and-vxo.

(28)

Tirole, Jean, and Roland Bénabou, 2010, Individual and Corporate Social Responsibility, Economica 77.

Wang, Kevin Q. and Xu, Jianguo, 2015, Market volatility and momentum, Journal of Empirical Finance 30, 79-91.

2018 Report on US Sustainable, Responsible and Impact Investing Trends, The Forum for Sustainable and Responsible Investment,

(29)

Appendix A

The ESG scores used are those developed by ‘Thomson Reuters’ and retrieved from DataStream. The scores are a composite percentile based on over 178 distinct measures subdivided between an Environmental, Social and Governance Score. The distribution of these measures across the total score and their respective ESG pillars are shown in Figure 1. Figure 1: The distribution of ESG (Environmental, Social and Governance) measures in the ‘Thomson Reuters’ Asset4 ESG score: “TRESGS”

This figure presents the distribution of 178 individual ESG measures used to determine the ESG score of an individual firm across main pillars and their subsequent distribution. Measure counts within the respective pillars or distributions are enclosed by parentheses. Data source: Thomson Reuters4

4_{https://www.refinitiv.com/content/dam/gl/en/documents/methodology/esg-scores-methodology.pdf} 34% 35% 31% ESG PILLARS

Environmental (61) Social (63) Governance (54)

33%

35% 32%

ENVIRONMENT MEASURES

Resource Use (20) Emissions (22) Innovation (19)

45%

13% 22%

20%

SOCIAL MEASURES

Workforce (29) Human Rights (8)

Community (14) Product Responsibility (12)

62% 23%

15%

GOVERNANCE MEASURES

(30)

Appendix B

Table 1: Descriptive Statistics for the Environmental, Social and Governance Score of North American and European firms between 2002 and 2017

This table presents the descriptive statistics (mean, minimum, maximum, median, standard deviation and observations) of portfolio constituent firms their Environmental, Social and Governance Score between 2002 and 2017. Descriptive Statistics are aggregated per year; underlying data is on a monthly basis. Source: DataStream.

Panel A: North America Good 2.5%

2002 81.02 77.57 85.27 81.04 2.57 96 2003 82.75 78.51 87.33 81.84 2.93 107 2004 83.39 79.21 91.94 83.92 2.83 163 2005 85.77 81.80 92.83 84.04 3.60 180 2006 85.58 81.37 95.96 84.87 3.84 181 2007 85.46 80.95 95.54 85.03 2.95 215 2008 85.89 82.45 96.71 85.26 2.59 279 2009 87.45 84.15 97.63 86.68 2.93 300 2010 87.76 84.21 97.63 87.47 2.94 324 2011 87.09 83.50 93.93 86.92 2.44 336 2012 85.49 82.11 91.75 85.18 2.71 336 2013 86.13 83.24 91.93 85.37 2.29 348 2014 86.05 81.53 94.23 85.26 3.29 372 2015 85.20 81.42 94.23 84.42 3.31 580 2016 85.95 81.60 93.65 85.03 2.89 794 2017 84.53 78.70 95.35 84.69 3.89 666 Total 85.67 77.57 97.63 85.25 3.33 5277

Panel B: North America Good 5%

(31)

Panel C: North America Good 10%

2002 73.91 66.20 85.27 72.44 5.07 384 2003 74.35 65.81 87.33 73.65 6.27 405 2004 74.97 67.05 91.94 72.10 5.96 630 2005 77.57 69.98 92.83 76.01 5.86 696 2006 78.90 71.43 95.96 77.82 4.83 713 2007 78.13 70.91 95.54 76.76 5.23 834 2008 79.48 72.76 96.71 78.79 4.58 1090 2009 80.87 73.96 97.63 80.51 4.89 1200 2010 81.22 74.75 97.63 80.84 4.77 1272 2011 81.16 75.13 93.93 80.45 4.24 1308 2012 79.99 74.32 91.75 79.15 3.97 1331 2013 80.31 74.15 91.93 79.90 4.24 1371 2014 79.50 73.51 94.23 77.96 4.60 1464 2015 78.32 71.31 94.23 76.93 4.93 2299 2016 77.91 69.83 93.65 76.67 5.67 3164 2017 74.99 65.77 95.35 73.34 6.68 2656 Total 78.44 65.77 97.63 77.93 5.64 20817

Panel D: North America Bad 10%

(32)

Panel E: North America Bad 5%

2002 21.37 0.00 27.74 25.22 8.73 192 2003 22.59 0.00 28.25 25.44 8.30 206 2004 24.52 0.00 30.86 27.32 9.05 319 2005 23.74 0.00 29.30 27.25 8.36 348 2006 21.97 0.00 28.50 24.56 7.90 361 2007 21.25 0.00 25.55 22.25 4.47 420 2008 20.36 0.00 23.98 21.83 4.18 549 2009 20.05 0.00 23.60 20.93 3.84 600 2010 20.40 0.00 25.01 20.89 4.34 636 2011 19.76 0.00 23.65 20.71 3.88 660 2012 19.23 0.00 23.39 20.77 4.32 671 2013 19.52 0.00 23.77 20.88 4.19 687 2014 19.94 0.00 23.48 20.71 3.61 737 2015 20.54 11.61 24.00 21.53 2.80 1152 2016 20.52 8.08 24.02 21.58 3.27 1587 2017 20.51 11.46 23.82 21.11 2.71 1331 Total 20.61 0.00 30.86 21.53 4.73 10456

Panel F: North America Bad 2.5%

(33)

Panel G: Europe Good 2.5%

2002 84.32 80.78 87.90 84.37 2.43 108 2003 81.13 78.93 84.22 80.16 1.73 115 2004 84.68 81.35 93.66 82.65 3.54 168 2005 87.06 82.41 95.99 86.72 3.50 192 2006 87.40 84.20 96.39 85.41 3.23 204 2007 85.17 82.08 94.39 83.75 3.02 216 2008 85.97 83.40 94.39 85.07 2.80 228 2009 88.50 84.67 95.68 87.89 2.76 240 2010 88.07 85.36 95.68 87.41 2.65 252 2011 88.64 86.24 95.15 87.90 2.38 264 2012 88.25 85.56 94.14 87.90 2.44 276 2013 88.37 85.40 95.11 87.50 2.59 276 2014 88.62 85.50 92.75 88.25 2.10 291 2015 89.74 86.67 94.62 89.38 2.26 324 2016 89.67 86.13 94.83 89.38 2.50 334 2017 90.34 86.99 95.97 89.63 2.39 277 Total 87.82 78.93 96.39 87.66 3.33 3765

Panel H: Europe Good 5%

(34)

Panel I: Europe Good 10%

2002 77.90 72.39 87.90 76.65 4.54 396 2003 76.63 72.35 84.22 76.33 3.29 438 2004 78.63 73.43 93.66 78.05 4.55 648 2005 79.84 72.54 95.99 78.69 5.16 756 2006 80.89 74.43 96.39 79.57 4.76 781 2007 80.19 75.25 94.39 79.45 3.62 840 2008 81.71 76.71 94.39 81.64 3.36 908 2009 82.85 77.50 95.68 82.31 4.00 948 2010 83.07 77.88 95.68 82.66 3.64 1005 2011 83.61 78.38 95.15 83.14 3.69 1044 2012 83.22 78.19 94.14 82.30 3.68 1068 2013 83.08 78.14 95.11 82.21 3.81 1080 2014 83.90 78.29 92.75 83.55 3.48 1149 2015 84.31 79.33 94.62 83.05 3.71 1268 2016 84.56 80.17 94.83 83.61 3.57 1306 2017 85.12 79.80 95.97 84.40 3.78 1088 Total 82.51 72.35 96.39 82.16 4.43 14723

Panel J: Europe Bad 10%

(35)

Panel K: Europe Bad 5%

2002 29.14 26.00 32.79 29.08 2.15 204 2003 28.31 21.21 32.67 29.66 3.32 222 2004 27.09 11.57 32.04 28.59 4.28 327 2005 26.62 17.91 30.60 27.96 3.31 384 2006 23.97 14.19 28.93 24.68 3.71 396 2007 22.35 12.19 27.02 23.09 3.45 420 2008 22.20 10.80 26.89 22.88 3.32 456 2009 22.95 11.43 27.39 24.16 3.82 480 2010 22.67 10.33 28.04 24.55 4.75 504 2011 22.01 9.55 27.85 22.84 4.14 528 2012 21.85 8.98 28.32 23.40 4.78 540 2013 22.63 6.66 29.33 24.40 5.21 540 2014 21.53 6.66 28.29 23.24 5.51 576 2015 22.89 10.57 28.14 24.18 4.07 636 2016 24.37 11.56 29.38 25.89 4.60 658 2017 24.92 13.46 31.38 26.42 4.46 547 Total 23.59 6.66 32.79 24.64 4.71 7418

Panel L: Europe Bad 2.5%

(36)

Appendix C

Table 1: Correlation matrix of all independent and dependent variables between 2002 and 2017

This table presents the correlations between portfolio returns of “Good”, “Bad” and “Good Minus Bad” (GMB). The common risk factors as control variables, the market risk premium (MRP), the size factor (SMB), the book-to-market factor (HML) and as the dependent variable volatility indices (VIX) between 2002 and 2017. Good”, “Bad” and “GMB” denote market value weighted portfolios of the top % ESG score, bottom % ESG score or the long top % short bottom % ESG score firms respectively. Underlying data is on a monthly basis.

Source: DataStream, Kenneth R. French5_.

(37)